Autocorrelation to detect "dominant frequency"?

[hdsjulian]

Code below.
I want to use a microphone, external I2S ADC and ESP32 to determine the note being played into a microphone. So i want to be able to sing or play an instrument into the mic and detect the dominant frequency.
Assuming the mic is used right, how would i go along with this? Currently the FrequencyDetector spits out frequencies all over the place. I'd also need a threshold to make sure that background noises aren't "analyzed" but using VolumeMeter i only get 32768 as value...
What am I doing wrong?

#include <Arduino.h>
#include <AudioTools.h>

// I2S pins
#define I2S_BCK_PIN 14   // Bit Clock (BCK)
#define I2S_LRCK_PIN 15  // Left-Right Clock (LRCK)
#define I2S_DATA_PIN 12  // Data (DOUT)

I2SStream i2sStream; //AudioInfo info(8000, 1, 16);
AudioInfo info(8000, 1, 16);
FrequencyDetectorAutoCorrelation out(1024);
StreamCopy copier(out, i2sStream); 


void setup() {
  Serial.begin(115200); // Initialize Serial for Serial Plotter
  AudioToolsLogger.begin(Serial, AudioToolsLogLevel::Warning);

  // Configure I2S using AudioTools
  auto cfg = i2sStream.defaultConfig(RX_MODE);
  cfg.bits_per_sample = 32;
  cfg.channels = 1; // Mono
  cfg.sample_rate = 96000;
  cfg.is_master = false; // Slave mode
  cfg.i2s_format = I2S_STD_FORMAT;
  cfg.use_apll = true;
  cfg.pin_bck = I2S_BCK_PIN;
  cfg.pin_ws = I2S_LRCK_PIN;
  cfg.pin_data_rx = I2S_DATA_PIN; // Correct pin for RX mode

  i2sStream.begin(cfg); // Start I2S
  out.begin(info);
}

void loop() {
  copier.copy(); // Copy data from I2S to output stream
  Serial.print("Frequency: ");
  Serial.println(out.frequency(0)); // Print detected frequency to Serial
}

[pschatzmann]

How do you think this can work if you provide data with the sample rate of 96000 and 32 bits, but you tell the FrequencyDetectorAutoCorrelation that it receives 16 bits at 8000 samples per second ?

The requency detector will work if you provide a single note and the correct information about the input signal: if you play multiple notes together, you should use FFT...

[pschatzmann]

You did not write what microphone you were using: you there might be a chance that 16 bits are not supported!

Please follow the advice given here and test with a sine wave.

A pure tone is a simple sound wave with a single frequency, like those produced by a tuning fork, while a human singing voice is a complex sound wave with multiple frequencies, including a fundamental frequency and harmonics, so you should rather use FFT.

There is a chapter about volume control in the Wiki...

[hdsjulian]

i'm using a SHURE SM 57 attached to an i2s adc board so i'd assume that 16 bits are very well supported. but then i'll test a FFT solution, thanks

[pschatzmann]

Visualize the audio data first, to confirm that you receive some valid sound...

[hdsjulian]

Ok. So i did visualize and thought everything was fine. Then i amended the code without committing it. Now i'm struggling to get a new visualization done which makes me think something else is wrong. Will get back to you. thanks for helping :)

[hdsjulian]

So somehow the ground between the ESP and the ADC was disconnected and thus the data was gibberish. Fixed.

Now i can get a very clear signal and some waveform when whistling into the mic. The AutoCorellation does not reflect that, however, just spitting out more or less arbitrary numbers.